Deep Fehler

Back in the old days there used to be German havannah bots like Deep Fork and Ring von Fehler. I wonder what happened to them.

Ah, sorry, still there. I invited Deep Fork to a base-10 game, only to find out that it hasn't outgrown base-8, apparently. So I invited it to a base-8 game, but as yet to no avail either.

So what about (quote)

_"Simply wait until late 2012: the bots will pull off your undertrousers early enough ;-) "

Ingo._

Ingo?

I should note that I can play base-10 sparring games against Wanderer and Castro anytime. In that context this quote is interesting:

_"You made your offer (in 2002) that no bot will beat you in a single game (out of ten games) within ten years. No programmer - at any time - made a public announcement that (s)he will give you sparring time before such a match."

Ingo_

And not only interesting (not only should I win ten, the opponent wants me to have an extra handicap, how sporting), but clearly wrong.

Apart from that, what does that mean anyway? The precise conditions were no issue in 2002. They have now been established by the organizers and they include my right to know 'who' I'm playing against by means of a couple of sparring games. I don't see why the 'programmer' should dictate the conditions.

I've made a very modest use of the opportunity to play against Wanderer and Castro, without having to to claim that as my 'right', in any way. Both programmers obviously agree it is a gentlemanly thing to do in the first place.

I bought an extra pair of undertrousers, don't tell me Deep Fork is bolting out! ;-)

Hi, Christian. You say "They [the precise conditions] have now been established by the organizers ..." Who are the organizers, when will it take place, and what are the conditions? I just might be interested. :)

Hi Richard,

Sorry, I thought Ton van der Valk

http://www.littlegolem.net/jsp/info/player.jsp?plid=19214

would have contacted you. The precise conditions concern computercrashes and the like. Other than that it's just life games with 1 hour each, presumably with 30 sec. + per move.

Omar Syed is also involved at the 'live coverage' side.

I need a guarantee of course that I'm playing against a bot. That I'm playing myself is easy to check.

Regarding technical hiccups, I don't really expect them, but if small problems are solved within the time of the side encountering the problem, I'm fine with that.

If a game must be paused altogether for technical reasons, I'm not sure what the procedure would be, so the precise conditions for that occasion are not known to me at this moment because of the other parties involved, i.e. you, Timo, and possibly others - I still don't see a game against Deep Fork. ;-)

I second Richard's question. Are the conditions published somewhere on the web? Can I link to them from the Havannah Google+ page or publish them there?

Should you consider Lajkonik worthy of playing in the match, he is not avoiding fight. I am reconstructing its engine to the state from October-November 2011, when he was able to play games like this one:

Game #1392056

Later, in preparation to the Computer Olympiad, I introduced lots of ad hoc changes without due diligence and testing. They have degraded the playing strength of Lajkonik. However, due to spring cleaning and my other duties, the reconstruction works do not proceed as fast as we might wish.

Menawhile, my mathematician friends Åukasz Lew and Witold Jarnicki are developing a novel paradigm of computational game-playing. If their theory crystallizes well, it may help machine brains overcome human ones.

I am looking forward to your reply.

Best wishes,

Marcin Ciura

Ah, it seems that writing the previous post took me too long and Christian answered in the meantime. Thank you!

Hi Marcin,

Of course Lajkonik is welcome. But I play at most 10 games total, so there will be a distribution problem. Because I play against different bots, I want an opportunity to familiarize myself with their style of play in a couple of sparring games, some time before the actual event.

Unfortunately I had to point out to one contestant who likes to see me blindfolded on top of being handicapped by the sudden death character of the event, that I will not play against a bot I don't know.

I'll ask Ton to get something on the web, and the Havannah Google+ page seems like a good place.

Your final remark is thrilling! Can you please elabotrate on it (as far as you can or may), preferably in a new thread?

Hi Christian,

Since there are more or less five bots up for the challenge, may I propose that each of them plays two games with you? I apologize if my words were unclear: as soon as Lajkonik is repaired, I will let him play here, also in a base-10 version.

As to the new theory, I will try to explain my grasp of its present state. I have not been doing serious mathematics for some twenty years so my understanding is more intuitive than formal. Please be patient while I am gathering my mind. Lukasz, who incidentally once counted among top Havannah players here, could explain it much better than me.

I'm all eager anticipation, although like you, my understanding is intuitive rather than formal. I really hope it is something that sheds more light on how humans play, because I can do it, yet not explain what I do. Like riding a bike or catching a ball for that matter. Why are simple things so complicated?

As to the distribution, if there are five bots, two games each would be perfect. Also, I'm quite willing to finish the whole match should I lose a game, just for the fun of it. :)

Hi everyone,

Omar Seyd started an off topic discussion at the Arimaa Forum on the rules on the Havannah Challenge 2012. Off course LG, mindsports or the Google+ Havannah page would seem more on topic. Nevertheless Omar deserves credits for asking the right questions:

http://arimaa.com/arimaa/forum/cgi/YaBB.cgi?board=other;action=display;num=1333125858

I resume the main questions:

- Who can participate in the challenge?

- How to deal with program-bugs (I remember this was an issue at the computer Olympiad in Tilburg in november 2011)

- How much thinking time?

- How many games a day?

- When playing on-line, both sides need impartial referees who can stay in contact by mail or phone. Both sides need to agree on the referees.

Anything else?

Come to think of it, two games a day would suit me perfectly. A nicely filled week in the summer :)

Marcin, you mentioned 'more or less five bots'. I take it that Castro, Wanderer and Lajkonik are included. Ingo Althoefer has personally informed me that he is "supporting a bot team", and that was the full stretch of info. So I assume it is Deep Fork version x.x.

Which one is #5?

If the theory by Lukasz and Witold develops well (we have started to describe it, please be patient), I will put up a program based on it. In fact, its first fruits have already been implemented (under the code name Antares). But I would not like to clog too many slots so I can withdraw the weaker of the two programs. Four bots can play a mini-tournament before the match to determine who gets two extra games.

Deep Fork apparently hasn't played for a year or so and in any case didn't yet accept my invitations. It is as Ingo corrected me in another thread (regarding the same subject):

_"Christian

Science sometimes is going strange ways, and people from outside (like you) do not understand the motivation."_

I'm not sure whether the implication is intended to be that the motivation is as hard to understand as the science. I suggested secrecy might be observed because the end justifies the means. This may underline the stupidity of 'people like me'. Something brilliant must be brewing in Germany.

I don't wish to sound negative, but you're not even in the top 10 ranked humans presently - are you planning on getting some anti-bot training from some of the real shining stars of the game before such an undertaking?

Dude. He is the inventor of the game, and was the first strong player. That is meaningful.

It was his challenge to make, and is his to defend. He can do whatever he sees fit to do.

If other players want to extend a similar challenge, they can.

I, also, don't want to sound too negative but speaking for "myself" I suspect that as of today anybody in the top 50 would be expected to (easily?) win the "Freeling Bet". However the optimist in me hopes for some serious improvements either here or from one of the other bots before the big match.

Also, depending on how many games I will actually play against Freeling, there are some thoughts of doing some anti-human tuning in an attempt to steal a win.

Castro has already won a few games against Christian. Christian is probably the favourite, but winning 10 out of 10 is not going to be a walk in the park, for any human.

Hello,

a few comments, not intended to run in a long debate.

(i) In the original announcement (from 2002) it was

not at all said that Christian had to be familiar with

his opponent. Especially, it was not required that

he knew details of the program(s) or the team(s) behind it/them.

(ii) Richard Lorenz wrote:

> Also, depending on how many games I will actually play against

> Freeling, there are some thoughts of doing some anti-human

> tuning in an attempt to steal a win.

You are not alone with this attitude.

(iii) Darse Billings wrote:

> Christian is probably the favourite, but winning 10 out of 10

> is not going to be a walk in the park, for any human.

Agreed.

Christian is clear favorite, even with taking point (i) seriously.

But street fighting is part of the business.

For instance, read Hsu's book on the 1997 match between Kasparov and

Deep Blue, or Jonathan Schaeffers book "one jump ahead" on Checkers.

Ingo.

The second Kasparov match was a bit testy, mostly because Kasparov is testy. (Murray had some interesting insider comments on that).

Marion was a consumate gentleman, and Jonathan did everything he could to accomodate his wishes, knowing that Marion was extending him a tremendous privilege. (When the checkers federation changed the rules after-the-fact to forbid a computer challenger for the world championship, Tinsley promptly resigned his long-standing title in order to allow the Man vs Machine match to go forward). Jonathan is very competitive, and always wants to win, but he has always held the utmost respect and admiration for Marion (appropriately so). [BTW, i did read One Jump Ahead, and helped proof-read it -- i especially like the parts where he quotes me.:]

In fact, most of the Man vs Machine matches (e.g. Othello, Logistello vs Murakami; Scrabble, Maven vs Adam Logan; Backgammon, TD_Gammon vs Malcolm Davis; Poker, Polaris vs Laak and Eslami, etc.) have been extremely friendly, positive events, and i see no reason why they shouldn't be. Both sides represent a tremendous level of accomplishment in the domain, and not in direct conflict with each other (one is mastering the game itself, the other is mastering the AI programming of the game). As a programmer, i'm always happy to accomodate whatever the human champion wishes, within reason. I don't think Christian is being at all unreasonable, and i doubt Timo thinks so either (he made Castro open source several months ago).

If a MvM match isn't fun and rewarding for both sides, maybe it isn't worth doing.

Darse wrote:

> The second Kasparov match was a bit testy,

A bit? It was full mode street fighting.

> mostly because Kasparov is testy. (Murray

> had some interesting insider comments on that).

Things started already in the first negotiations after

the 1996 match. Kasparov came to Yorktown Heights

and asked for modified conditions for the 1997 match.

"Allow me a notebook with three types of software:

(i) An opening data base.

(ii) Endgame data bases.

(iii) An engine like Fritz for tactical blunderchecks."

IBM said "no, no, no". And more: From that moment on "they"

knew about Kasparows missing self-confidence.

> Marion was a consumate gentleman, and Jonathan did

> everything he could to accomodate his wishes, knowing

> that Marion was extending him a tremendous privilege.

Agreed. When I mentioned Jonathan's very readable book

"One jump ahead" I did not mean the special relation between

Jonathan and Marion, but the atmosphere in the early

tournament years of Chinook.

> In fact, most of the Man vs Machine matches ...

> have been extremely friendly, positive events,

Hmm, I would prefer to say "some of them".

> ... Scrabble, Maven vs Adam Logan;

Perhaps that match. But in the doctoral dissertation of

Brian Sheppard (author of Maven) episodes from the early

days of Scrabble bots are described where human players

used special tricks to keep the upper hand.

> ... and i see no reason why they shouldn't be.

> ... I don't think Christian is being at all unreasonable,

Sometimes I have a memory like an elephant.

Between 2002 and 2009 Christian was in big mouth

mode, concerning computer chances in Havannah. Read the

original announcement from 2002, or his early postings

in 2009 here in the forum.

Then, gradually he switched into more polite gears.

This is his good right. But he can not demand that others

switch in the same way he has done.

To make things clear:

I definitely prefer fair sportmanship. See my experiences

in the matches with 3-Hirn chess. I learned a lot

about fair play in competition with chess masters and

grandmasters and about "Sitzfleisch" and "Turnierhaerte".

Ingo.

If a MvM match isn't fun and rewarding for both sides, maybe it

isn't worth doing.

"Between 2002 and 2009 Christian was in big mouth mode, concerning computer chances in Havannah. Read the original announcement from 2002, or his early postings in 2009 here in the forum."

Oh boy, if MCTS hadn't emerged programmers would still be grabbling in the dark. I have no point to prove, I just needed some 'big mouth mode' to even get some attention for the fact that yes, Chess bots were at grandmaster level, but hey, what about a pencil and paper game and 8yo could learn in a minitue or so? A few years back Ingo had trouble seeing Havannah as a touchstone for programming in the first place.

How does the human brain work? What exactly is 'pattern recognition'? How come bots suck at Havannah but can be made invincible in Explocus, and without much effort at that. How come humans can look deeply in some games and not at all at others that are as simple? Why are some games so easy to program and others so difficult?

Those are my interests, not to 'prove' anything in particular about Havannah. It's just a good vehicle for asking those questions.

I agree that "if a MvM match isn't fun and rewarding for both sides, maybe it isn't worth doing". A condescending and elitist attitude about people that are not scientists (like me) doesn't really support 'fun and reward'. Your genius is occupied with my game, remember? And I'd rather fight on the board.

(i) Richard Lorenz wrote:

> Also, depending on how many games I will actually play against Freeling, there are some thoughts of doing some anti-human tuning in an attempt to steal a win.

(ii) Ingo wrote:

You are not alone with this attitude.

In the above context of gaining more insight, this ad-hoc attitude of 'stealing a win' seems a bit out of place. It's not what this is about in my opinion. But you're welcome.

> Dude. He is the inventor of the game, and was the first strong player. That is meaningful.

Well, it's a sentence that parses - congratulations.

> It was his challenge to make, and is his to defend. He can do whatever he sees fit to do.

Nor do I say otherwise. I just think that if even Kasparov can surround himself with expert advisors before a high profile match against a computer, I think it's wise for anyone to.

> If other players want to extend a similar challenge, they can.

Nor do I say otherwise.

Were you attempting to actually address something that I did say?

I will :)

"I don't wish to sound negative, but you're not even in the top 10 ranked humans presently - are you planning on getting some anti-bot training from some of the real shining stars of the game before such an undertaking?"

No, but your use of the adjective 'real' is suggestive. I never considered myself a top player. I'm an inventor and the 'wiring' needed for that differs from a player's. But I'll give it my best shot.

As for the opposition, consider it an extra handicap: the bots are represented by the best programmers, humans by mediocrisy. Mediocricy to win ten out of ten. Any more wishes?

I mean, in Chess it's Carlsen, Anand, Kramnik, whoever against King_of_the_Chess_bots, and it's about who wins the match, not who wins every game.

That difference is what it's about and what justifies the questions about how humans evaluate a position and how bots evaluate a position and why the results are so different and sometimes so couterintuitive.

Humans use concepts like a 'plan' and 'pattern recognition' - what are they exactly?

Some simple games turn out to be hard to program beyond 'human level' (Hex, Go, Havannah), while others are relatively easy (Explocus, Othello). Why can humans look deep into a Go position but not into an Explocus position. Is it just the columns? Why then can humans look deep into an Emergo or Bashni position and not into an Othello position?

Hurraying a bot that can compete with Carlsen is all very fine, and solving Checkers was a milestone, but it doesn't answer these questions.

Two comments.

(1) Re your last post, Christian, unfortunately it is unlikely this match, or what is learned as a result of it, will get us any closer to answering your big questions. As you point out, the algorithm of choice for Havannah bots is MCTS and I can say with a fair degree of confidence that no human uses anything similar when playing a game. Hopefully we'll learn a little something about improving AI, but I don't think we will learn anything about HI (Human Intelligence).

(2) There is the following exchange, edited below:

(i) Richard Lorenz wrote:
... there are some thoughts of doing some anti-human tuning in an attempt to steal a win.

(ii) Ingo wrote:
You are not alone with this attitude.

(iii) Christian wrote: In the above context of gaining more insight, this ad-hoc attitude of ‘stealing a win’ seems a bit out of place. It’s not what this is about in my opinion. But you’re welcome.

(iv) I write: As mentioned above, I don't believe we are gaining the kinds of insights you are hoping for, anyway. But more directly to the point, my comment was a reaction to Fat Phil's question to Christian: "...are you planning on getting some anti-bot training?" Since most humans have some kind of (anti-bot) strategy when playing bots, it seems fair for bots to do the same. Call it opponent modelling?

Yes, and I'm fine with it. However, I'm not immediately inclined to agree with your assumption that humans 'don't use anything similar' as MCTS.

I don't know what humans use. I know they name it, but that's the human condition: first we name it, later we try to find out what it is.

So what is a 'plan' or what is 'pattern recognition'? I'm fairly sure that the way I look at a Havannah position is at least not less similar to MCTS than the traditional evaluation based on certain characteristics of a position. Patterns are dynamic and seeing a continuation is sifting through 'randomness', taking most of it out of the equation at first sight.

"Hopefully we’ll learn a little something about improving AI, but I don’t think we will learn anything about HI (Human Intelligence)."

To avoid misunderstandings, I agree. I see this match as part of raising those 'bigger questions', not answering them. :)

When is the matches played and can they be watched live? Would be a lot of fun.

Omar Syed of Arimaa and the organizers are working on that. Other than Havannah, Arimaa has been designed with the goal of being accessible for humans but difficult for bots. That's one question I'm interested in: what is needed to make a game accessible for humans and difficult for bots. And Arimaa and Havannah are two very different examples without much common ground. So the answer would appear to be quite elusive.

You forgot to mention that another goal of Arimaa was, that it's playable with a standard chess set. Otherwise you just could have created a connection game on a large board and bots will have problems.

Right, and a good feature too. Regarding 'large boards', that's a variable I would very much like to be kept in check, so to say. ;-)

Christian: What's your opinion about games on infinite size boards?

Or with unbounded size (finite but can grow unlimitedly)?

Just out of curiosity, Christian, who are these organizers (of the challenge match) you are referring to?

I think it's pretty clear what kind of games are amenable to computer search, and what kinds harmonize well with the slow but massively parallel pattern-matching systems of human brains. Mind you, i've been thinking about that topic and playing with both sides for the last 20 years... ;-)

Nebulous goals versus concrete (easily measured) goals is one example of an important distinguishing feature. Large branching factor is another. The details are a bit too much to summarize here though. Maybe over a beer sometime.

@Ypercube
I start thinking about them and then remember I stopped. As far as my games go Hexade can be played on any size upwards from crammed while remaining basically the same game. I've never used it as a determining a priori characteristic.

@Richard
Ton van der Valk (Tony - http://www.littlegolem.net/jsp/info/player.jsp?plid=19214) of www.hexboard.com is the main organizer. You can easily contact him.

@Darse
Good subject, I'm glad to find someone with some expertise in this area :) But later, it's midnight over here.

I think it is a really interesting challenge and really looking forward to seeing it :)

Likewise. Even though I'm a programmer, I'm cheering for the human, because I don't like the way far too often "throw more CPU at it" is proposed as a solution to a problem. At the moment, a doubling, or even octupling, of computational efficiency shouldn't be enough to put the top computer above the top human, which is a situation I'm happy with.

However, as a programmer, I look forward to algorithmic breakthroughs that eventually put the computer on top.

If the only advances are massive parallelism over clusters, then quite frankly I'm not interested, and the programmers have lost, even if they can win the game.

I agree with FatPhil.

I often heard the term pattern matching/recognition here, I think another important aspect is 'divide and conquer'. The humans are able to divide the problem in sub-problems and then combine those sub-solutions to a whole plan. Especially in connection games this is true. But those tasks are not easy for a computer: how to divide the problem? What if the sub-problems overlap with each other? What do the sub-solutions tell us about the global solution? And many more questions could be asked. So another strength of the humans is to divide the problem and combine the sub-solutions _with respect to the global solution_.

Another technique I sometimes use is to think that I need "something somewhere over there". I can then think about developments elsewhere on a board without actually precisely defining what and where that something or somewhere are. It's like a quantum superposition of states. After a while the 'something'may become important and I will see that some of the choices are better than others, and the quantum superposition collapses onto something precise. So I looked at maybe half a dozen potential moves with only one roll-out.

I've never seen any computer program use this kind of fuzzy approach.

"If the only advances are massive parallelism over clusters, then quite frankly I'm not interested, and the programmers have lost, even if they can win the game."
Or even one game, in this particular case.

_"I think it's pretty clear what kind of games are amenable to computer search, and what kinds harmonize well with the slow but massively parallel pattern-matching systems of human brains. Mind you, i've been thinking about that topic and playing with both sides for the last 20 years... ;-)

Nebulous goals versus concrete (easily measured) goals is one example of an important distinguishing feature. Large branching factor is another. The details are a bit too much to summarize here though."_

There are imo. three basic variables regarding programmability of abstract two-player perfect information zero-sum games.

# The games, with all their different themes and mechanisms.
# The different human approaches to different games.
# The different program approaches to different games.

That's a lot of variables already to attempt to say something general. For instance, the number of excellent Chess, Shogi, XiangQi, Draughts and Go players dwarfs the number of excellent players of newer games. That's inherent, but it clouds a more general comparison of what games are difficult to program up to 'human level'. Human level in Havannah is not the same as human level in Go because a high summit needs the support of a broad base. In that light the result of one particular challenge may be interesting, but hardly enough to draw general conclusions.

You mention the branching factor. Arimaa was designed not to just have a large branching factor, but one that allows humans to read a position deeper than a bot could manage. The question is still open, but the human side should always be taken into account if a game is to be 'hard for bots'.

More later :)

The human brain is differently suited for different types of games. To consider this, it doesn't seem very useful to seperate themes and mechanics too emphatically, because only a _combination_ emerges as more or less difficult for the human brain (or for a bot for that matter). If we look at an arbitrary and far from complete thematic division:

# Chess variants
# Elimination games
# Territory games
# Connection games

Can anything at all be said about differences in difficulty in programming games belonging to these classes? The question is not rethorical. Chess variants, in the widest sense, all provide a focuspoint in the presence of an absolute piece. The eyes of a master travel over a position with the king as constant point of departure and return. It gives the brain a good handle it seems. Draughts games are very different in this respect. Marion Tinsley commented on this by saying that playing Chess was like looking over an infinite ocean, and playing Checkers was like looking in a bottomless pit. Edgar Allen Poe commented on it in 'The Murders in the Rue Morgue': "I will therefore take occasion to assert that the higher powers of the reflective intellect are more decidedly and more usefully tasked by the unostentatious game of draughts than by all the elaborate frivolity of chess."

Does the human mind sees Chess-type and Draughts-type games differently. Does the fact that 'Chess-type' is so dominated by Chess itself give the Chess bots an advantage? Will International Draught go the way of Checkers? If so, why does it take so long? Or will Draughts bots remain weaker than the best players, suggesting that Draughts is 'more difficult to program'? Does Draught's large margin of draws, compared to Chess make a difference? Or the effort involved?

I'm not even considering territory and connection yet :)

Looking at territory and connection, the mechanics are usually different from chess and draughts-type games, the main shift being from movement to placement. Of course there are exceptions. Corey Clark's Slither combines movement and placement as does LOA, Claude Soucie's dynamic connection game that even involves capture.

Regarding 'the state of affairs', as far as I know Othello bots rule, while Go bots suffer. Both may be considered to have top human opposition.

Hex and Havannah bots still seem to suffer, but human opposition may not yet be up to par.

There's a shift in method. MCTS, based on random play-outs, appears to give a better foundation than the traditional evaluation function and search procedures in Go and Havannah, presumably in Hex, but strangely enough not in Othello. The best Othello bots use the traditional method. Since they already outplay human competition, the incentive to use a MCTS based method may be low.

Intuitively one would consider MCTS a dead-end street in Chess and Draughts-type games, or in Arimaa for instance, but a promising approach in placement games. At least that's how I feel it. Any insights why this should be the case, if indeed it is?

Intuitively, too, one would consider placement games easier to program, but that doesn't seem to be the case just yet. Or is it just the effort that goes into Chess and Draughts programming? Quite a few programmers are involved in Go programming too.

Alpha-beta search is an exhaustive shallow search, combined with an evaluation function that attempts to capture "knowledge" about the position with a few salient features. It was clear to me 20 years ago that that was rather one-dimensional, and had been given far too much attention due to its fabulous success in chess. That's more of a comment on chess (a highly tactical game with good locality of reference) than on the algorithm. [See the McCarthy quote on "very fast fruit flies".]

I came up with a different algorithm (called depth budgeted search, DBS) to explore much lighter and deeper, which is essential in some domains. [That was the basis of my SameGame solver, which set all-time high score records on all the sites i used it on. It plays *waaay* beyond the level of human play, but that's mostly trivia, based on how many humans play the game seriously. I'm more interested in how well the game can be played, and less interested in the human-centric view of everything.] UCT is basically the same idea, with automatic control of search allocation (mine did on-the-fly learning).

Monte Carlo / UCT had a dramatic impact in computer Go, which had been languishing for 30 years. Lots of Go players had smugly concluded that no computer could ever play "their" game, simply because it hadn't happened yet. Typical humans.

The past is not a good predictor of the future. Don't underestimate what computers can do. There is nothing special about wetware -- computer algorithms can do all of that, and a whole lot more.

@Darse

I just saw a newsitem about players moving pieces with their brain, their head connected to various sensors. The game wasn't the essence, that was the communication framework for severely handicapped people. But there's still a lot of interest in wetware and how it works :) .

I'm interested in the human component for several reasons. Shortly the new CodeCup Challenge will be announced. The game to program will be Symple.

So I asked myself the question what would happen. Thirty or forty programs compete and one will win, eventually, no special gifts of prophecy required.

Such a result doesn't mean one iota regarding the game's 'resistance' to programmability if there's no human opposition to provide a context. The SameGame solver is new to me (does it play Havannah?) as is DBS or the idea behind UCT (a refinement of MCTS as far as I understand), but the fact that it plays way beyond human level is a context you also readily provide.

So I started playing Symple against 'Marcel Vlastuin' who let's himself represent by a MCTS based Symple-bot. This is what makes it interesting.

Symple was discovered rather than designed and there was no intention to implement any characteristics that would make it more resistent to programming it. The game is indeed simple in that a premiss of three points makes the rules self explanatory. In other words: Symple is quintessential. Since we're in a Hex/Havannah forum, Hex is also quintessential, but Havannah is just a lucky merger of three winning structures, two of which are arbitrary. Havannah's design was based on choices.

The Symple premiss is a thematic merger of territory and 'dynamic connection' on a placement basis:

\* Get one point for every stone
\* Lose P points for every group
\* On your turn, either grow every group or start a new one

There's no capture and moving is compulsory. A singleton is a group by definition. P is an even number upwards from 4. The territory component is clear, the 'dynamic connection' is embedded in the group penalty: connect two groups and win P points.

So Symple is as interesting as it gets, programwise. Simple rules and structure, while multiple placements per turn explode the branching factor. Yet MCTS already has led to a program that, for all means and purposes, is about as strong as I am, at the moment. But I'm getting better (initially I lost most games, now I'm winning most).

We cannot know how the best programs would score against the best humans, because small numbers mean nothing. I weep for a humanity that would have me as it's reperesentative in a Symple human versus bot contest. But we need a lot of different games and a lot of games played in each of them, to have some insight in where bots rule and where humans still hold their ground :) .

bc. In 2002 Freeling offered a 1000 Euro prize for any computer program in ten years that could beat him in even one game in a 10-game match.

--Wikipedia, almost verbatim

That sounds like each program gets 10 games, 10 chances to win against Christian.

If instead there are to be a grand total of 10 games, what if there are 11 or more challengers? And how is the prize to be divided among the programs, if more than one, that win at least one of the ten games?

On the pragmatic side, as it looks now, there will be something like 5 programs and all programmers are on friendly terms and flexible enough. How the price will be divided isn't my problem and considering its size it will probably be no problem for the contestants either. It's not that it would solve anyone's mortgage problems, if any :) .

Also, I'm willing to play 10 games regardless of a loss.